Playing with AI_Can AI help creating ferrous discrimination?

[Tinkerer]

Here is today's chat Today.docx

[moodz]

Moodz, yes it is funny how the Chat GTP is lazy and gives me just the first type of answer it finds on line. Only when I insist and pressure it, it digs deeper.
I have to learn how to ask the right questions.
It seems the Chat GTP realises that I work on an intuitive level based on personal observations, so it answers to me on a similar level.

I think you are using copilot ? Try Claude ... he/it is more analytical and focussed on giving you an actual solution to an answer if you ask the right question.
You can drag and drop a ltspice circuit .asc file onto the chat bar and Claude will read it.
He can modify them as well.

Copilot is more like one of those salesmen in a second hand car lot.

[Tinkerer]

moodz, thank you for the suggestion to use Claude. That sounds much better.

Copilot is more like one of those salesmen in a second hand car lot. Great analogy!

[Aziz]

Hi Tinkerer,

moodz, thank you for the suggestion to use Claude. That sounds much better.

Copilot is more like one of those salesmen in a second hand car lot. Great analogy!

try Google AI too. And compare the results. It would be interesting to see, what is really better.
I try it on Google AI first before making the same AI query on Claude AI. Google AI has no limitations compared to Claude AI.
Aziz

[moodz]

Hi Tinkerer,



try Google AI too. And compare the results. It would be interesting to see, what is really better.
I try it on Google AI first before making the same AI query on Claude AI. Google AI has no limitations compared to Claude AI.
Aziz

Not sure what limitations you are talking about..

Google AI only appears to be more proficient on the surface .. but when actually delivering the goods ( working production code ... Claude is a killer ).
Google AI is like the fresh new guy from college .. full of ideas but shallow on delivery. Claude is like the seasoned engineer .. solid production ready code.

Below is the google AI comparison of Claude to Gemini Pro ( AKA google ) ... even Google admits Claude is better.
I have tried both and Claude leaves the other platforms in the weeds. Dont worry ... I will swap as soon as any are better than Claude.

This is what Google says ...

Claude 4.6 Opus is generally considered superior for serious coding, complex debugging, and agentic tasks due to its higher reliability and superior reasoning, while Google Gemini 3 Pro excels in speed, integration within IDEs, and tasks needing broad context. Claude holds a slight edge in accuracy, but Gemini offers better value through broader ecosystem integration.


Claude 4.6 Opus (Anthropic) for Coding

• Best for: Complex refactoring, deep reasoning, and high-stakes coding tasks.
• Strengths: Often seen as more reliable with complex instructions, resulting in fewer errors and better "agentic" capabilities for managing multi-step tasks.
• Performance: Consistently ranked higher for production-ready code generation.
• Workflow: Excellent at handling long-context, repository-level analysis.

Gemini 3 Pro (Google) for Coding

• Best for: Fast prototyping, integrated development, and searching for current information.
• Strengths: Highly integrated into Google ecosystem and IDEs, with strong "agentic" speed, though sometimes less accurate than Claude.
• Performance: Comparable to Claude on benchmarks (e.g., SWE-bench), but sometimes requires more intervention on complex tasks.
• Workflow: Stronger when you need real-time data or a quick all-in-one assistant.

Key Comparison Areas

• Accuracy: Claude is widely regarded as more dependable for generating working code on the first try.
• Complexity: Claude shines in, navigating complex codebases, while Gemini is excellent at rapid, smaller tasks.
• Speed & Cost: Gemini 3 Flash is often faster, but Claude tends to have better long-term efficiency by requiring less manual correction.

Ultimately, for deep, complex, and high-reliability coding, Claude 4.6 Opus is the preferred tool, while Gemini is excellent for faster, integrated tasks.

[Aziz]

Hi Paul,

Not sure what limitations you are talking about..

this applies for non-payers. You can't use it as much as you want it for free. There is a limit. If this is reached, you have to wait till you can go on.
Aziz

[moodz]

Hi Paul,



this applies for non-payers. You can't use it as much as you want it for free. There is a limit. If this is reached, you have to wait till you can go on.
Aziz

OK here is an idea. Use Google AI to hack Claude so you can use it.

[Aziz]

OK here is an idea. Use Google AI to hack Claude so you can use it.

Comeon Paul,

you could do the Hi-Q mono coil VLF with your 5 bucks board with ease. Just drive the coil with a rectangular pulse coming from an IO-line. You don't need much driving power (a few mA only).
I would be interested on your depth results.
Aziz

[moodz]

Comeon Paul,

you could do the Hi-Q mono coil VLF with your 5 bucks board with ease. Just drive the coil with a rectangular pulse coming from an IO-line. You don't need much driving power (a few mA only).
I would be interested on your depth results.
Aziz

which mono VLF are you using ... there are several designs.

[Aziz]

which mono VLF are you using ... there are several designs.

The very very basic High-Q LC-Tank driving TX.
You can pulse the TX directly out of the logic level IO output line (3.3 V I think in your case). We don't need much TX power. Only few mA. The IO-line should be able to drive the TX. High-Q LC-tanks do not need high power.
But imagine, you can do it with a pure mono coil.

Below is the simple LC-tank. And a zipped LTspice file.

Aziz

PS: Forget the RX-Coil part in the circuit above. Single demod of the TXref is enough.

[moodz]

Aziz, Sorry mate but I have already been down this path. You cant operate on the Q curve of a resonant circuit and expect to extract any target information at depth.
Below is a picture of a test "ground" ferrite and a 1 gram nugget. The ferrite will "pull" your resonant coil changing both amplitude and phase ... and the change in Q makes it even worse.
and then you expect to extract the target signal from that huge ferrite signal and overcome the 6 power return loss from TX -> target -> RX. No chance.

In air you could make it work ... but any ground or conductive losses like salt are going to cause serious problems.
Realistically you are trying to pull say 100 nanvolts out of many volts of noise + target.

I do have a suggestion though .... use double sideband suppressed carrier modulation at the TX so that the coil is operated with the LSB below resonance and the USB above resonance.
shifts in the LSB and USB will balance out ...by comparing addition and subtraction of the relevant parameters ( phase and amplitude in the USB / LSB separately ) you can resolve targets whilst overcoming resonance drift.
It still sucks though as a method.

[Aziz]

Hi Paul,

..
It still sucks though as a method.

yes, for small gold in extreme mineralized ground, that's for sure. I agree with you.
The X-response is 1000+ times more and masking the R-resonse heavy. The GB projection method does not deliver enough signal strength for tiny targets on extreme ground.
But I am surprised, that the basic mono coil VLF can still be realised as a low/mid-level detector. A 10x10 cm thin Al-foil can be detected at approx. 70 cm (with GB).

Only the true IB configuration can deliver the required signal strength. At least if we push more current through the TX coil or using an RX-amplifier.

What is your AI mate proposing for good gold detector on heavy mineralized ground? This would be quite interesting. For us all.
My AI mate sucks.

Cheers,
Aziz

[moodz]

Hi Paul,



yes, for small gold in extreme mineralized ground, that's for sure. I agree with you.
The X-response is 1000+ times more and masking the R-resonse heavy. The GB projection method does not deliver enough signal strength for tiny targets on extreme ground.
But I am surprised, that the basic mono coil VLF can still be realised as a low/mid-level detector. A 10x10 cm thin Al-foil can be detected at approx. 70 cm (with GB).

Only the true IB configuration can deliver the required signal strength. At least if we push more current through the TX coil or using an RX-amplifier.

What is your AI mate proposing for good gold detector on heavy mineralized ground? This would be quite interesting. For us all.
My AI mate sucks.

Cheers,
Aziz

umm the Cayley transform detects a 10x10 cm foil at a meter ( for a 40 cm coil ) or coil 2.5 diameters ...and I have not hit the noise floor with it yet.
This is in the same league as PI detectors for sensitivity ... how can it be so good ?.... it does not require a phase rotation.
Its essentially a time domain process so its similiar to PI ... you transmit then you analyse the captured RX signal.
The AI is all over it and the maths .. mainly because the AI I use really really likes maths.
A step impulse is a whole lot of stacked synchronous sine waves .... PI generates a step impulse, VLFs generate sine waves ...
Now do you see where this is going ?

PS I have a broadband mono coil design that perfectly balances and supports PI or VLF modes .... but I cant tell you about it yet till the maths is worked out.

[Aziz]

Hi Paul,

I have added some responses to my simple mono coil VLF project in my thread. The GB is the standard projection of the response to the ground response vector. It works pretty well. But not for tiny gold nuggets in hot ground.
It's a pretty simple nice toy however.
Aziz

[Tinkerer]

So the Claude AI is very helpful. It offered to help me with the preamp. I showed it this screenshot and it offered to transform it into an LTspice file so I could simulate it. It sent me this back. Interesting. Why is it like that? I am going to ask it the why s. Curious to learn from the answers.